SingleImageSR use case

The Single Image Super Resolution (SISR) use case is build to compare the image quality between different SiSR solutions. A SiSR algorithm inputs one frame and outputs an image with greater resolution. These are the methods that are being compared in the use case:

  1. Fast Super-Resolution Convolutional Neural Network (FSRCNN) [Ledig et al., 2016]
  2. Single Image Super-Resolution Generative Adversarial Networks (SRGAN) [Dong et al., 2016]
  3. Multi-scale Residual Network (MSRN) [Li et al., 2018]
  4. Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) [Wang et al., 2018]
  5. Content Adaptive Resampler (CAR) [Sun & Chen, 2019]
  6. Local Implicit Image Function (LIIF) [Chen et al., 2021]

A use case in IQF usally involves wrapping a training within mlflow framework. In this case we estimate quality on the solutions offered by the different Dataset Modifiers which are the SISR algorithms. Similarity metrics against the Ground Truth are then compared, as well as predicted Quality Metrics.

Execution

The number of runs are all the combinations between repetitions, modifiers list as well as hyper parameter changes.

(you can skip this step in demo pre-executed datasets)

Metrics

ExperimentInfo is used to retrieve all the information of the whole experiment. It contains built in operations but also it can be used to retrieve raw data for futher analysis. Its instance can also be used to apply metrics per run. Some custum metrics are presented. They where build by inheriting Metric from iq_tool_box.metrics.

(you can skip this step in demo pre-executed datasets)

Noise and Sharpness (Blind) Metrics

Regressor Quality Metrics

All Metrics Comparison